(1) Quantifying Opportunity-Adjusted Risk Taking (OART) in Football Pass Selection

Abstract. This paper proses Opportunity-Adjusted Risk Taking (OART), a new statistic for quantifying passing risk preferences in football. OART measures risk by comparing a player's chosen pass receiver against all available teammates at the moment of the pass, using predicted completion probabilities from a contextual model. We analyze 68,515 pass events from the 2022 FIFA World Cup using StatsBomb 360° tracking data, retaining 56,415 events (82.3%) for final analysis across 64 matches and 608 players. Our gradient boosting pass success model achieves AUC = 0.886 ± 0.006 and Brier score = 0.091 ± 0.002. OART demonstrates moderate reliability (split-half r = 0.558, 95% CI [0.482, 0.634]) with systematic individual differences (player-level range 0.122-0.770). The metric correlates strongly with completion rate (r = -0.73) as theoretically expected for a risk measure.

Status. Exploratory working paper. Significant refinements to be made.

PDF

(2) Structural Constraints on Information Aggregation in Directed Networks

Abstract. This paper investigates how the topological structure of directed networks constrains the set of outcomes achievable under local, decentralized information dynamics. Rather than asking what equilibrium a particular learning rule will reach, we ask a prior question: what outcomes are structurally impossible given only the network topology, regardless of the specific update rule employed? Through a combination of theoretical analysis and computational simulation across multiple network structures and update rules, we identify conditions under which consensus is structurally precluded and characterize how the decomposition of a directed graph into strongly connected components shapes the space of possible asymptotic states. Results suggest that certain impossibility results are robust across a broad class of local averaging dynamics.

Status. Working paper. Feedback welcome.

PDF

(3) Sensitivity of Network Statistics to Graph Construction Methods in Reddit Communities

Abstract. This paper examines the sensitivity of commonly reported directed network statistics to graph construction choices in Reddit interaction data. Using networks from 10 subreddits across 11 months (110 network pairs), we compare chain-based construction against reply-based construction. We compute density, reciprocity, largest strongly connected component fraction, in-degree Gini coefficient, and PageRank statistics for each construction. Results show strong rank correlations between constructions (rho = 0.875-1.000) but systematic differences in metric magnitudes. Chain networks exhibit higher density but lower reciprocity and degree inequality than reply networks. These findings demonstrate that network construction choices substantially affect metric values while preserving relative ordering, with implications for comparative studies in online social networks.

Status. Methodological note. Basis for ongoing research.

PDF